How TDSQL Dominated Global OLAP & OLTP Benchmarks: Inside the Technical Secrets

Benchmark Milestones

In the latest TPC‑DS decision‑support benchmark, Tencent Cloud's TDSQL scored 72.6 million QphDS with a price/performance ratio of 37.52 CNY/kQphDS, surpassing the previous world record by 282% and reducing cost by 37%. Earlier, TDSQL also topped the TPC‑C transaction‑processing benchmark, demonstrating dominance in both OLAP and OLTP workloads.

TPC‑DS is a global standard that simulates a large‑scale retail data‑analysis scenario, running 99 distinct analytical queries to evaluate a database’s query latency, concurrency, and throughput.

Three Core Technical Capabilities

The record‑breaking results stem from three self‑developed technologies that optimize distributed interaction, parallel resource utilization, and single‑core performance.

1. MPP‑Based Distributed Interaction

TDSQL implements a custom Massive Parallel Processing (MPP) execution framework that replaces the traditional pull‑based data request model with a push‑based approach. Each node proactively pushes required data to downstream nodes, eliminating bottlenecks and fully decoupling query shards, which enables seamless, high‑speed execution of all tasks.

2. Parallel Execution Framework

The parallel execution engine removes inter‑process blocking, raising CPU utilization above 95%. Fine‑grained task decomposition and dynamic resource scheduling ensure every CPU core is fully leveraged. The engine combines data parallelism, task parallelism, and pipeline parallelism, dramatically accelerating large‑scale data processing.

3. Columnar Storage & Vectorized Execution

By storing data column‑wise, TDSQL improves compression and read efficiency. The vectorized engine leverages SIMD instructions to process multiple data items per instruction, delivering multiple‑fold gains in single‑core throughput. Deep algorithmic and data‑structure optimizations align the engine with modern CPU architectures.

Implications and Outlook

These innovations demonstrate that achieving double‑digit performance gains while cutting costs requires more than hardware upgrades; it demands fundamental redesign of data movement, execution parallelism, and storage layout. TDSQL’s success illustrates a roadmap for future database systems aiming for world‑class performance in both analytical and transactional workloads.